sprint

Sprint

Autonomous multi-agent development plugin for Claude Code. Spec-driven, iterative sprints with specialized agents.

Part of Agentic Forge — Claude Code plugins for autonomous AI workflows.

Stop prompting in circles. Sprint replaces ad-hoc AI coding with structured, specification-driven development. Write specs, run /sprint, and let coordinated agents handle the rest.

At its core, the /sprint command is a spec-driven, self-iterative state machine — it reads your specifications, orchestrates specialized agents through defined phases, and loops autonomously until the work is done or validation passes.

What is Sprint?

Sprint is a Claude Code plugin that turns Claude into an autonomous development team:

• Project Architect analyzes requirements, creates specifications, and coordinates work
• Implementation Agents (Python, Next.js, CI/CD, or any tech via allpurpose-agent) build features according to specs
• Testing Agents (QA, UI) validate the implementation
• Sprint Orchestrator — the self-iterative state machine that manages phases, handoffs, and convergence

You write specs → Agents implement → Tests validate → Iterate until done

The orchestrator drives the loop: specs in, working code out. No manual intervention required between phases.

Why It Works

Unlike single-shot prompting where context bloats and AI mistakes compound, Sprint uses a convergent multi-pass approach:

• Context preservation — Each agent receives only what it needs (specs, contract, relevant code). No wasted tokens on irrelevant history.
• Specs shrink, not grow — Completed work is removed from specs. Each iteration focuses only on what remains.
• Errors get erased — Working code stays untouched while issues get fixed. The signal-to-noise ratio improves with each pass.

Think of it like a diffusion process: the picture starts noisy, but with each iteration, the noise reduces and clarity emerges. By the final pass, only the solution remains.

Most sprints converge well before 5 iterations. If they don't, the system pauses and asks you what to do — adjust specs, continue iterating, or intervene manually. You stay in control.

Multi-Paradigm Design

Sprint is technology-agnostic. While it includes specialized agents for Python/FastAPI and Next.js, the system works with any tech stack:

• The allpurpose-agent adapts to Go, Rust, Flutter, Ruby, or any technology
• Create your own specialized agents for your preferred stack
• The architect automatically selects appropriate agents based on project structure

The Second Brain Effect

Two files give agents persistent memory across sprints — reducing token usage and keeping context focused:

.claude/project-goals.md — The business brain (you maintain this)

• Product vision and target audience
• Market analysis and differentiators
• Success metrics and constraints
• What you're building and why

The more detail you provide, the sharper and more shrewd the architect becomes.

.claude/project-map.md — The technical brain (architect maintains this)

• Project structure and architecture
• API surface and database schema
• Routes, components, environment variables
• Where everything lives and how it connects

Agents read this instead of scanning the entire codebase. The architect keeps it lean and current.

Installation

From Agentic Forge (recommended)

# Add the marketplace
/plugin marketplace add damienlaine/agentic-forge

# Install the plugin
/plugin install sprint

# Update to latest version
/plugin marketplace update damienlaine/agentic-forge

Local Development

# Clone this repo
git clone https://github.com/damienlaine/agentic-sprint.git

# Run Claude Code with the plugin
claude --plugin-dir ./agentic-sprint

Quick Start

1. Set Up Your Project

# Interactive project onboarding
/sprint:setup

This creates both Second Brain documents through guided questions:

• .claude/project-goals.md (business vision)
• .claude/project-map.md (technical architecture)

2. Create Your First Sprint

/sprint:new

This creates .claude/sprint/1/specs.md. Edit it with your requirements.

3. Run the Sprint

/sprint

Watch the agents work:

1. Architect analyzes specs and creates detailed specifications
2. Implementation agents build in parallel
3. Testing agents validate the work
4. Architect reviews and iterates (up to 5 times)
5. Sprint completes with a status summary

Commands

Command	Description
/sprint	Run the full sprint workflow
/sprint:new	Create a new sprint
/sprint:setup	Interactive project onboarding
/sprint:test	Manual UI testing with live browser
/sprint:generate-map	Generate project-map.md
/sprint:clean	Remove old sprint directories

Manual Testing Mode